The aim of the project is to make a brain-inspired model of object recognition.
The model’s kernel will be applicant’s quantum associative (neural-like) nets.
Although many artificial object-recognition methods have been developed, it is nevertheless fruitful to try an approach that systematically imitates human vision taking into account as much neurobiological data as possible. When we observe
an object from various viewpoints, its images are different (view-dependent), but we anyway consider it as the same object —
since all the corresponding images converge into the same view-independent attractor. Hebbian learning categorizes the inputs into roughly orthonormal eigen-images which act as attractors.
Additionally, ICA-like infomax pre-processing, which produces Gabor wavelets, would ensure that edges and contours
are indeed detected as the essential ingredients of eigen-images – thus defining objects which are subsequently
recognized by attractor-net processing. Such expectations are supported by simulations of Bartlett & Sejnowski.
In sum: An integrative model will be made where infomax-produced image-coding Gabor wavelets are used as pre-processed inputs to the quantum associative net which then recognizes the objects.